Some additional reading on statistical analysis of RNA-seq: http://www.nathalievialaneix.eu/doc/pdf/tutorial-rnaseq.pdf

# pacman is a good package for loading packages
library(pacman)
pacman::p_load(edgeR, RColorBrewer, gplots, ggplot2, reshape2, DT, cowplot,
               limma, DESeq, DESeq2, data.table, e1071, ComplexHeatmap)

Source R script

I have writen several functions in an external R script, we can use those function by sourcing the function

source("https://raw.githubusercontent.com/CJinnny/RNA-seq-tutorial/master/RNA_seq_tutorial_functions.R")

Data preprocessing

This data came out of Kallisto software, gene counts are the sum of trancript pseudocounts. These counts are not in round numbers, so we need to round them to integers first.

Alternatively, you should be able to generate count data using HTSeq https://htseq.readthedocs.io/en/release_0.10.0/ or featurecounts http://bioinf.wehi.edu.au/featureCounts/. Please do not use bedtools as it usually counts ambiguous alignments.

Here I created my own transcriptome assembly using StringTie https://ccb.jhu.edu/software/stringtie/, hense the “MSTRG” gene_id in the count data.

The reason I did that was because I not only want to study genes found in the reference genome, but also characterize new transcripts.

In order to know gene functions, we can convert MSTRG id into maize reference genome id (v4), which is already well characterized.

require(data.table)
data <- fread("https://raw.githubusercontent.com/CJinnny/RNA-seq-tutorial/master/genes.txt", sep = "\t")
data[,6:15] = round(data[,6:15])

id_conv <- fread("https://raw.githubusercontent.com/CJinnny/RNA-seq-tutorial/master/https://raw.githubusercontent.com/CJinnny/RNA-seq-tutorial/master/all_MSTRG_v4_v3.csv", header = TRUE, sep = ",")
id_conv <- data.frame(id_conv, row.names = 1)
functions <- fread("https://raw.githubusercontent.com/CJinnny/RNA-seq-tutorial/master/B73v4_gene_function.txt", header = FALSE, sep = "\t")
functions <- subset(functions, select = c("V1", "V2"))
colnames(id_conv) <- c("v4_gene_id", "v3_gene_id", "MSTRG_gene_id")
colnames(functions) <- c("v4_gene_id", "gene_function")
annotations = merge(id_conv, functions, by = "v4_gene_id", all.x = TRUE)
data <- merge(data, annotations, by = "MSTRG_gene_id", all.x = TRUE)
data <- aggregate(data, 
                  by = list(data$MSTRG_gene_id),
                  FUN = aggregate_func
                  )
data <- na.omit(data, cols="v4_gene_id")
data <- data.frame(data, row.names = 1)
setnames(data, old = c("UU1", "UU2", "UU3", "UU4", "UU5", "WW1", "WW2", "WW3", "WW4", "WW5"), 
         new = c("U1", "U2","U3","U4","U5", "W1", "W2","W3","W4","W5"))
head(data)
##             MSTRG_gene_id chr_name    start      end strand  U1  U2  U3
## MSTRG.1000     MSTRG.1000     chr1 25047512 25050032      - 195 333 154
## MSTRG.10002   MSTRG.10002    chr10 32585811 32588982      - 154 244 273
## MSTRG.10005   MSTRG.10005    chr10 32815479 32817141      +   0   0   0
## MSTRG.10007   MSTRG.10007    chr10 32529623 32567490      - 419 381 397
## MSTRG.1001     MSTRG.1001     chr1 25152267 25155144      + 139 152 150
## MSTRG.10011   MSTRG.10011    chr10 33226451 33367505      - 609 523 514
##              U4  U5  W1  W2  W3  W4  W5     v4_gene_id    v3_gene_id
## MSTRG.1000  166 105 117  52 126 123  81 Zm00001d028167 GRMZM2G003506
## MSTRG.10002 106 242 226 212 274 218 212 Zm00001d023969 GRMZM2G425986
## MSTRG.10005   0   0   0   0   0   1   0 Zm00001d023973 GRMZM2G097286
## MSTRG.10007 223 395 423 345 406 435 375 Zm00001d023968 GRMZM2G124974
## MSTRG.1001   81 190 167  92 110 182 148 Zm00001d028171 GRMZM2G071374
## MSTRG.10011 274 595 508 320 421 375 399 Zm00001d023974 GRMZM2G027431
##                                                            gene_function
## MSTRG.1000           Cytochrome b561 and DOMON domain-containing protein
## MSTRG.10002                         A/G-specific adenine DNA glycosylase
## MSTRG.10005                     Cysteine proteinases superfamily protein
## MSTRG.10007                                    Glutathione S-transferase
## MSTRG.1001  Calcium-dependent lipid-binding (CaLB domain) family protein
## MSTRG.10011                  Putative endonuclease or glycosyl hydrolase

Is the data homoskedastic?

counts <- data[,6:15]

require(e1071)
data_diagnosis(counts)
## skewness is:
##  36.63689 41.82896 21.72463 34.99034 22.78793 42.0665 33.39914 37.64356 27.01962 54.97658 
## kurtosis is:
##  2199.034 3028.689 865.9139 1964.686 867.6066 2758.378 1949.117 2319.712 1375.454 4894.567

Data transformation

Data is skewed (high skewness and high kurtosis).

We need to do data transformations for PCA plot, otherwise outliers will have a great impact on the clustering.

log transformation: logcounts = log2(counts + 1).

DESeq2::rlog(): “regularized log” transformation. For more information see https://rdrr.io/bioc/DESeq2/man/rlog.html

edgeR::cpm(): “counts per million” transformation. For more information see https://rdrr.io/bioc/edgeR/man/cpm.html

DESeq2:varianceStabilizingTransformation(): “variance stabilizing transformation”. For more information see https://rdrr.io/bioc/DESeq2/man/varianceStabilizingTransformation.html

require(DESeq2)
require(edgeR)
logcounts = log2(counts + 1)
rlogcounts = rlog(as.matrix(counts))
rownames(rlogcounts) = rownames(logcounts)
cpmcounts = cpm(as.matrix(counts), prior.count = 2, log = TRUE)
vstcounts = varianceStabilizingTransformation(as.matrix(counts))
data_diagnosis(logcounts)
## skewness is:
##  -0.5997057 -0.6048998 -0.6093726 -0.4912253 -0.6109366 -0.5945748 -0.5262748 -0.5730369 -0.5900014 -0.552007 
## kurtosis is:
##  -0.7826695 -0.7977504 -0.8067969 -0.9002022 -0.7725521 -0.7831297 -0.9279888 -0.8513029 -0.8382193 -0.8962603

data_diagnosis(rlogcounts)
## skewness is:
##  -0.6679657 -0.6836715 -0.6832402 -0.6763476 -0.6779365 -0.6773912 -0.6746331 -0.6697096 -0.6842158 -0.671056 
## kurtosis is:
##  -0.6403723 -0.6370364 -0.6450905 -0.6332838 -0.633821 -0.6331829 -0.6479403 -0.6440739 -0.6427632 -0.6518892

data_diagnosis(cpmcounts)
## skewness is:
##  -0.4285089 -0.4730039 -0.4773158 -0.4526601 -0.4639736 -0.4572871 -0.4528065 -0.4374913 -0.4839089 -0.438765 
## kurtosis is:
##  -0.9539685 -0.9381164 -0.9576858 -0.9321906 -0.9269013 -0.9280453 -0.9964032 -0.9797621 -0.9533996 -1.003599

data_diagnosis(vstcounts)
## skewness is:
##  0.3270751 0.2803435 0.27878 0.3042513 0.3098411 0.308709 0.2888468 0.312236 0.2699977 0.2917919 
## kurtosis is:
##  -0.5191474 -0.5951802 -0.6988962 -0.5030264 -0.550126 -0.5454259 -0.6343581 -0.5475308 -0.6730192 -0.6279794

Draw PCA (Principle Component Analysis) plot

require(graphics)
require(RColorBrewer)
par(mfrow=c(2,3), mar=c(5.1, 4.6, 4.1, 1.6))
draw_PCA(counts, title = "PCA on raw data")
draw_PCA(logcounts, title = "PCA on log transformed data")
draw_PCA(rlogcounts, title = "PCA on rlog transformed data")
draw_PCA(cpmcounts, title = "PCA on cpm transformed data")
draw_PCA(vstcounts, title = "PCA on vst transformed data")

Draw MDS (multidimensional scaling) plot

par(mfrow=c(2,3))
plotMDS(counts, col = c(rep("red", 5), rep("blue", 5)), cex = 1.5)
title("MDS plot on raw data")
plotMDS(logcounts, col = c(rep("red", 5), rep("blue", 5)), cex = 1.5)
title("MDS plot on log transformed data")
plotMDS(rlogcounts, col = c(rep("red", 5), rep("blue", 5)), cex = 1.5)
title("MDS plot on rlog transformed data")
plotMDS(cpmcounts, col = c(rep("red", 5), rep("blue", 5)), cex = 1.5)
title("MDS plot on cpm transformed data")
plotMDS(vstcounts, col = c(rep("red", 5), rep("blue", 5)), cex = 1.5)
title("MDS plot on vst transformed data")

Draw correlation heatmap

require(gplots)
draw_corr_heatmap(as.matrix(counts), show_cellnote = TRUE, title = "clustering sample-to-sample\n distance on raw data")

draw_corr_heatmap(as.matrix(logcounts), show_cellnote = TRUE, title = "clustering sample-to-sample\n distance on log transformed data")

draw_corr_heatmap(rlogcounts, show_cellnote = TRUE, title = "clustering sample-to-sample\n distance on rlog transformed data")

draw_corr_heatmap(cpmcounts, show_cellnote = TRUE, title = "clustering sample-to-sample\n distance on cpm transformed data")

draw_corr_heatmap(vstcounts, show_cellnote = TRUE, title = "clustering sample-to-sample\n distance on vst transformed data")

Set conditions for DE analysis

Here I have to relevel the conditions because by default it considers the first factor to be the reference

conditions = factor(c(rep("Ufo",5), rep("Wt",5)))
conditions = relevel(conditions, ref="Wt")

Filter low-count reads

I’m using cpm normalization method, filter out non-expressed genes (0 counts), you can also try more stringent filtering (e.g. cpm > 1 in at least 5 samples)

keep <- rowSums(cpm(counts)>0) >= 5
table(keep)
## keep
## FALSE  TRUE 
##  3584 23106
keep_true = data.frame(keep[which(keep == TRUE)])
filter = subset(counts, rownames(counts) %in% rownames(keep_true))

DESeq

require(DESeq)
cds = newCountDataSet(filter, conditions)
cds = estimateSizeFactors(cds)
cds = estimateDispersions(cds)
DESeq_res = nbinomTest(cds, "Wt", "Ufo")
rownames(DESeq_res) = DESeq_res$id
DESeq_DE = subset(DESeq_res, (log2FoldChange < -1 & padj < 0.05) | (log2FoldChange > 1 & padj < 0.05) )
DESeq_nc = counts(cds, normalized = TRUE)
DESeq_nc = data.frame("id" = rownames(DESeq_nc), DESeq_nc)
DESeq_nc_DE = subset(DESeq_nc, id %in% DESeq_DE$id)

DESeq2

require(DESeq2)
colData = data.frame(samples = colnames(filter), conditions = conditions)
dds = DESeqDataSetFromMatrix(countData = filter, colData = colData, design = ~conditions)
## converting counts to integer mode
dds = DESeq(dds)
## estimating size factors
## estimating dispersions
## gene-wise dispersion estimates
## mean-dispersion relationship
## final dispersion estimates
## fitting model and testing
DESeq2_res = results(dds)
DESeq2_res = data.frame("id" = rownames(DESeq2_res), DESeq2_res)
DESeq2_DE = subset(DESeq2_res, (log2FoldChange < -1 & padj < 0.05) | (log2FoldChange > 1 & padj < 0.05) )
DESeq2_nc = counts(dds, normalized = TRUE)
DESeq2_nc = data.frame("id" = rownames(DESeq2_nc), DESeq2_nc)
DESeq2_nc_DE = subset(DESeq2_nc, id %in% DESeq2_DE$id)

edgeR

require(edgeR)
group = as.vector(conditions)
dge = DGEList(counts = filter, group = group)
dge = calcNormFactors(dge)
dis = estimateCommonDisp(dge)
tag = estimateTagwiseDisp(dis)
etx = exactTest(tag, pair = c("Wt", "Ufo"))
edgeR_res = etx$table
edgeR_res$FDR = p.adjust(edgeR_res$PValue, method = "BH")
edgeR_res = data.frame("id" = rownames(edgeR_res), edgeR_res)
edgeR_DE = subset(edgeR_res, (logFC < -1 & FDR < 0.05) | (logFC > 1 & FDR < 0.05) )
edgeR_nc = tag$pseudo.counts
edgeR_nc = data.frame("id"=rownames(edgeR_nc), edgeR_nc)
edgeR_nc_DE = subset(edgeR_nc, id %in% edgeR_DE$id)

limma

require(limma)
design = model.matrix(~conditions)
voom = voom(filter, design, normalize="quantile")
fit = lmFit(voom, design)
fit = eBayes(fit)
limma_res = topTable(fit, coef = NULL, n = Inf)
## Removing intercept from test coefficients
limma_res = data.frame("id" = rownames(limma_res), limma_res)
limma_DE = subset(limma_res, (logFC < -1 & adj.P.Val < 0.05) | (logFC > 1 & adj.P.Val < 0.05) )
limma_nc = 2**voom$E
limma_nc = data.frame("id" = rownames(limma_nc), limma_nc)
limma_nc_DE = subset(limma_nc, id %in% limma_DE$id)

Compare DESeq, DESeq2, edgeR and limma DEG results

dflist <- list(DESeq=DESeq_DE, DESeq2=DESeq2_DE, edgeR=edgeR_DE, limma=limma_DE)
Compare <- join_id(dflist)
Compare$message
## [1] "There are 282 genes DE in all 4 methods 114 genes in 3, 245 genes in 2, 136 genes in 1."

Displaying table using DT library

require(DT)
summary <- merge(Compare$merged_table, annotations, by.x = "id", by.y = "MSTRG_gene_id", all.x = TRUE)
datatable(summary,
          rownames = FALSE,
          caption = htmltools::tags$caption(
            style = "caption-side: bottom; text-align: left;",
            "Table 1: Summary of DEGs found in four different methods."),
          options = list(
            searchHighlight = TRUE
          )
          )

Add log2FoldChange

summary <- merge(summary, DESeq_res[,c("id", "log2FoldChange")], by = "id")
setnames(summary, old = "log2FoldChange", new = "DESeq_logFC")
summary <- merge(summary, DESeq2_res[,c("id", "log2FoldChange")], by = "id")
setnames(summary, old = "log2FoldChange", new = "DESeq2_logFC")
summary <- merge(summary, edgeR_res[,c("id", "logFC")], by = "id")
setnames(summary, old = "logFC", new = "edgeR_logFC")
summary <- merge(summary, limma_res[,c("id", "logFC")], by = "id")
setnames(summary, old = "logFC", new = "limma_logFC")
summary = summary[,c(1,9,10,11,12,6,7,8)]
summary = aggregate(summary, by = list(summary$id), FUN = aggregate_func)
summary = summary[,c(1,3,4,5,6,7,8,9)]
setnames(summary, old = "Group.1", new = "id")
datatable(summary, filter = "top",
          rownames = FALSE,
          caption = htmltools::tags$caption(
            style = "caption-side: bottom; text-align: left;",
            "Table 2: Summary of logFC values of DEGs."),
          options = list(
            autoWidth = TRUE,
            searchHighlight = TRUE,
            scrollX = TRUE
            )
          )

Draw Venn Diagram

require(VennDiagram)
draw_venndiagram(dflist, Compare$merged_table)

Draw heatmap for DEGs in separate plots

Cluster for Up & Down DEGs, respectively

require(ComplexHeatmap)
Ht1 = DE_heatmap(DESeq_nc_DE, title = "DESeq", km = 2)
Ht2 = DE_heatmap(DESeq2_nc_DE, title = "DESeq2", km = 2)
Ht3 = DE_heatmap(edgeR_nc_DE, title = "edgeR", km = 2)
Ht4 = DE_heatmap(limma_nc_DE, title = "limma", km = 2)
Ht1

Ht2

Ht3

Ht4

Draw heatmaps side-by-side using common DEGs

Ht1 = DE_heatmap(DESeq_nc_DE, common_id=Compare$common_id, title = "DESeq", km = 2)
Ht2 = DE_heatmap(DESeq2_nc_DE, common_id=Compare$common_id,  title = "DESeq2", km = 2)
Ht3 = DE_heatmap(edgeR_nc_DE, common_id=Compare$common_id, title = "edgeR", km = 2)
Ht4 = DE_heatmap(limma_nc_DE, common_id=Compare$common_id, title = "limma", km = 2)

Ht1 + Ht2 + Ht3 + Ht4

Draw MA plot

par(mfrow=c(2,2), mar=c(5.1, 4.6, 4.1, 1.6))
draw_MA(DESeq_res, type="DESeq")
draw_MA(DESeq2_res, type="DESeq2")
draw_MA(edgeR_res, type="edgeR")
draw_MA(limma_res, type="limma")

Draw Volcano plot

par(mfrow=c(2,2), mar=c(5.1, 4.6, 4.1, 1.6))
draw_volcano(DESeq_res, type="DESeq")
draw_volcano(DESeq2_res, type="DESeq2")
draw_volcano(edgeR_res, type="edgeR")
draw_volcano(limma_res, type="limma")

sessionInfo()
## R version 3.5.1 (2018-07-02)
## Platform: x86_64-apple-darwin15.6.0 (64-bit)
## Running under: macOS  10.14
## 
## Matrix products: default
## BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
## LAPACK: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libLAPACK.dylib
## 
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
## 
## attached base packages:
##  [1] grid      stats4    parallel  stats     graphics  grDevices utils    
##  [8] datasets  methods   base     
## 
## other attached packages:
##  [1] VennDiagram_1.6.20          futile.logger_1.4.3        
##  [3] plyr_1.8.4                  ComplexHeatmap_1.20.0      
##  [5] e1071_1.7-0                 data.table_1.11.8          
##  [7] DESeq2_1.22.1               SummarizedExperiment_1.12.0
##  [9] DelayedArray_0.8.0          BiocParallel_1.16.0        
## [11] matrixStats_0.54.0          GenomicRanges_1.34.0       
## [13] GenomeInfoDb_1.18.1         IRanges_2.16.0             
## [15] S4Vectors_0.20.1            DESeq_1.34.0               
## [17] lattice_0.20-38             locfit_1.5-9.1             
## [19] Biobase_2.42.0              BiocGenerics_0.28.0        
## [21] cowplot_0.9.3               DT_0.5                     
## [23] reshape2_1.4.3              ggplot2_3.1.0              
## [25] gplots_3.0.1                RColorBrewer_1.1-2         
## [27] pacman_0.5.0                edgeR_3.24.0               
## [29] limma_3.38.2               
## 
## loaded via a namespace (and not attached):
##  [1] colorspace_1.3-2       rjson_0.2.20           class_7.3-14          
##  [4] rprojroot_1.3-2        circlize_0.4.5         htmlTable_1.12        
##  [7] XVector_0.22.0         GlobalOptions_0.1.0    base64enc_0.1-3       
## [10] rstudioapi_0.8         bit64_0.9-7            AnnotationDbi_1.44.0  
## [13] splines_3.5.1          geneplotter_1.60.0     knitr_1.20            
## [16] Formula_1.2-3          jsonlite_1.5           annotate_1.60.0       
## [19] cluster_2.0.7-1        shiny_1.2.0            BiocManager_1.30.4    
## [22] compiler_3.5.1         backports_1.1.2        assertthat_0.2.0      
## [25] Matrix_1.2-15          lazyeval_0.2.1         formatR_1.5           
## [28] later_0.7.5            acepack_1.4.1          htmltools_0.3.6       
## [31] tools_3.5.1            bindrcpp_0.2.2         gtable_0.2.0          
## [34] glue_1.3.0             GenomeInfoDbData_1.2.0 dplyr_0.7.8           
## [37] Rcpp_1.0.0             gdata_2.18.0           crosstalk_1.0.0       
## [40] stringr_1.3.1          mime_0.6               gtools_3.8.1          
## [43] XML_3.98-1.16          zlibbioc_1.28.0        scales_1.0.0          
## [46] promises_1.0.1         lambda.r_1.2.3         yaml_2.2.0            
## [49] curl_3.2               memoise_1.1.0          gridExtra_2.3         
## [52] rpart_4.1-13           latticeExtra_0.6-28    stringi_1.2.5         
## [55] RSQLite_2.1.1          genefilter_1.64.0      checkmate_1.8.5       
## [58] caTools_1.17.1.1       shape_1.4.4            rlang_0.3.0.1         
## [61] pkgconfig_2.0.2        bitops_1.0-6           evaluate_0.12         
## [64] purrr_0.2.5            bindr_0.1.1            htmlwidgets_1.3       
## [67] labeling_0.3           bit_1.1-14             tidyselect_0.2.5      
## [70] magrittr_1.5           R6_2.3.0               Hmisc_4.1-1           
## [73] DBI_1.0.0              pillar_1.3.0           foreign_0.8-71        
## [76] withr_2.1.2            survival_2.43-1        RCurl_1.95-4.11       
## [79] nnet_7.3-12            tibble_1.4.2           crayon_1.3.4          
## [82] futile.options_1.0.1   KernSmooth_2.23-15     rmarkdown_1.10        
## [85] GetoptLong_0.1.7       blob_1.1.1             digest_0.6.18         
## [88] xtable_1.8-3           httpuv_1.4.5           munsell_0.5.0